Hi Kumaran, Have you tested Lustre using fileop. Is there any documentation present on using Fileop? can you specify the folder where you want to run Fileop? or will I have to copy Fileop to /mnt/lustre and try it there? Thank You Priya -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20060720/99d38989/attachment.html
Priya,
The fileop source file contains the necessary documentation.
Briefly, it is described below. You need to copy the benchmark to Lustre
FS (/mnt/lustre/) specifically to the directory (customized or default
striping conf) for which you would like to assess the metadata
performance.
This test is a metadata intensive
test. It is an N^3 engine that specifically tests
the metadata operations and performance.
Example:
fileop 10
This will create 10 directories, each of which will contain
10 directories, each of which will contain 10 files. Then
metadata operations will be performed on these 1000 files.
Example:
fileop 100
This will create 100 directories, each of which will contain
100 directories, each of which will contain 100 files. Then
metadata operations will be performed on these 1 million files.
Example output from fileop 2:
--------------------------------------
| Fileop |
| $Revision: 1.18 $ |
| |
| by |
| |
| Don Capps |
--------------------------------------
mkdir: Dirs = 6 Total Time = 0.032553911 seconds
Avg mkdir(s)/sec = 184.31 ( 0.005425652 seconds/op)
Best mkdir(s)/sec = 3858.61 ( 0.000259161 seconds/op)
Worst mkdir(s)/sec = 32.52 ( 0.030753851 seconds/op)
rmdir: Dirs = 6 Total Time = 0.001103878 seconds
Avg rmdir(s)/sec = 5435.38 ( 0.000183980 seconds/op)
Best rmdir(s)/sec = 7410.43 ( 0.000134945 seconds/op)
Worst rmdir(s)/sec = 2898.62 ( 0.000344992 seconds/op)
create: Files = 8 Total Time = 0.000985622 seconds
Avg create(s)/sec = 8116.70 ( 0.000123203 seconds/op)
Best create(s)/sec = 9532.51 ( 0.000104904 seconds/op)
Worst create(s)/sec = 6026.30 ( 0.000165939 seconds/op)
write: Files = 8 Total Time = 0.001550436 seconds
Avg write(s)/sec = 5159.84 ( 0.000193805 seconds/op)
Best write(s)/sec = 6442.86 ( 0.000155210 seconds/op)
Worst write(s)/sec = 4148.67 ( 0.000241041 seconds/op)
close: Files = 8 Total Time = 0.000099897 seconds
Avg close(s)/sec = 80082.18 ( 0.000012487 seconds/op)
Best close(s)/sec = 127100.12 ( 0.000007868 seconds/op)
Worst close(s)/sec = 27060.03 ( 0.000036955 seconds/op)
stat: Files = 8 Total Time = 0.000183582 seconds
Avg stat(s)/sec = 43577.18 ( 0.000022948 seconds/op)
Best stat(s)/sec = 91180.52 ( 0.000010967 seconds/op)
Worst stat(s)/sec = 10433.59 ( 0.000095844 seconds/op)
access: Files = 8 Total Time = 0.000101566 seconds
Avg access(s)/sec = 78766.27 ( 0.000012696 seconds/op)
Best access(s)/sec = 144631.17 ( 0.000006914 seconds/op)
Worst access(s)/sec = 23172.95 ( 0.000043154 seconds/op)
chmod: Files = 8 Total Time = 0.000674963 seconds
Avg chmod(s)/sec = 11852.50 ( 0.000084370 seconds/op)
Best chmod(s)/sec = 33554.43 ( 0.000029802 seconds/op)
Worst chmod(s)/sec = 3153.61 ( 0.000317097 seconds/op)
readdir: Files = 4 Total Time = 0.000214100 seconds
Avg readdir(s)/sec = 18682.87 ( 0.000053525 seconds/op)
Best readdir(s)/sec = 38479.85 ( 0.000025988 seconds/op)
Worst readdir(s)/sec = 7681.88 ( 0.000130177 seconds/op)
link: Files = 8 Total Time = 0.000689983 seconds
Avg link(s)/sec = 11594.48 ( 0.000086248 seconds/op)
Best link(s)/sec = 16131.94 ( 0.000061989 seconds/op)
Worst link(s)/sec = 4928.68 ( 0.000202894 seconds/op)
unlink: Files = 8 Total Time = 0.000451565 seconds
Avg unlink(s)/sec = 17716.17 ( 0.000056446 seconds/op)
Best unlink(s)/sec = 25575.02 ( 0.000039101 seconds/op)
Worst unlink(s)/sec = 6944.21 ( 0.000144005 seconds/op)
delete: Files = 8 Total Time = 0.001273155 seconds
Avg delete(s)/sec = 6283.60 ( 0.000159144 seconds/op)
Best delete(s)/sec = 7307.15 ( 0.000136852 seconds/op)
Worst delete(s)/sec = 4346.43 ( 0.000230074 seconds/op)
Hope this helps,
-Kums
>>> "Priya Muralidharan"
<Priya_Muralidharan@symantec.com> 7/20/2006
4:40 PM >>>
Hi Kumaran,
Have you tested Lustre using fileop. Is there any documentation
present on using Fileop? can you specify the folder where you want to
run Fileop?
or will I have to copy Fileop to /mnt/lustre and try it there?
Thank You
Priya
Priya,
Yes, I have used fileop on Lustre. The options for fileop may
be seen (along with a brief description of the test) by typeing
fileop with no options.
Yes, fileop tests the current working directory.
Enjoy,
Don Capps
----- Original Message -----
From: Priya Muralidharan
To: lustre-discuss@clusterfs.com
Sent: Thursday, July 20, 2006 5:40 PM
Subject: [Lustre-discuss] I/O performance on small files
Hi Kumaran,
Have you tested Lustre using fileop. Is there any documentation present on
using Fileop? can you specify the folder where you want to run Fileop?
or will I have to copy Fileop to /mnt/lustre and try it there?
Thank You
Priya
------------------------------------------------------------------------------
_______________________________________________
Lustre-discuss mailing list
Lustre-discuss@clusterfs.com
https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20060720/daebacb4/attachment.html
Kumaran,
Thank you for trying to help out.
Early versions of fileop are exactly what Kumaran described.
More recent versions now have options. To see the options
type fileop (with no options)
The current version of fileop (on the Iozone web site):
/*
* Author: Don Capps
* 3/13/2006
* Copyright: (2006)
* You are free to distribute and use this benchmark, for free.
* You are not permitted to distribute modified versions, or
* borrow code from this project to create proprietary products
* or competitive benchmarks. Any product that contains any
* of this code will be considered a derivative work and will
* be the sole property of me.
*
fileop [-f X ]|[-l # -u #] [-s Y] [-t] [-v] [-e] [-b] [-w]
-f # Force factor. X^3 files will be created and removed.
-l # Lower limit on the value of the Force factor.
-u # Upper limit on the value of the Force factor.
-s # Optional. Sets filesize for the create/write.
-t # Verbose output option.
-v # Version information.
-e # Excel importable format.
-b Output best case
-w Output worst case
*
* X is a force factor. The total number of files will
* be X * X * X ( X ^ 3 )
* The structure of the file tree is:
* X number of Level 1 directorys, with X number of
* level 2 directories, with X number of files in each
* of the level 2 directories.
*
* Example: fileop 2
*
* dir_1 dir_2
* / \ / \
* sdir_1 sdir_2 sdir_1 sdir_2
* / \ / \ / \ / \
* file_1 file_2 file_1 file_2 file_1 file_2 file_1 file_2
*
* Each file will be created, and then 1 byte is written to the file.
*
*/
---------
Enjoy,
Don Capps
----- Original Message -----
From: "Kumaran Rajaram" <krajaram@lnxi.com>
To: <lustre-discuss@clusterfs.com>; "Priya Muralidharan"
<Priya_Muralidharan@symantec.com>
Sent: Thursday, July 20, 2006 5:52 PM
Subject: Re: [Lustre-discuss] I/O performance on small files
> Priya,
>
> The fileop source file contains the necessary documentation.
> Briefly, it is described below. You need to copy the benchmark to Lustre
> FS (/mnt/lustre/) specifically to the directory (customized or default
> striping conf) for which you would like to assess the metadata
> performance.
>
> This test is a metadata intensive
> test. It is an N^3 engine that specifically tests
> the metadata operations and performance.
>
> Example:
> fileop 10
>
> This will create 10 directories, each of which will contain
> 10 directories, each of which will contain 10 files. Then
> metadata operations will be performed on these 1000 files.
>
> Example:
> fileop 100
>
> This will create 100 directories, each of which will contain
> 100 directories, each of which will contain 100 files. Then
> metadata operations will be performed on these 1 million files.
>
> Example output from fileop 2:
>
> --------------------------------------
> | Fileop |
> | $Revision: 1.18 $ |
> | |
> | by |
> | |
> | Don Capps |
> --------------------------------------
>
> mkdir: Dirs = 6 Total Time = 0.032553911 seconds
> Avg mkdir(s)/sec = 184.31 ( 0.005425652 seconds/op)
> Best mkdir(s)/sec = 3858.61 ( 0.000259161 seconds/op)
> Worst mkdir(s)/sec = 32.52 ( 0.030753851 seconds/op)
>
> rmdir: Dirs = 6 Total Time = 0.001103878 seconds
> Avg rmdir(s)/sec = 5435.38 ( 0.000183980 seconds/op)
> Best rmdir(s)/sec = 7410.43 ( 0.000134945 seconds/op)
> Worst rmdir(s)/sec = 2898.62 ( 0.000344992 seconds/op)
>
> create: Files = 8 Total Time = 0.000985622 seconds
> Avg create(s)/sec = 8116.70 ( 0.000123203 seconds/op)
> Best create(s)/sec = 9532.51 ( 0.000104904 seconds/op)
> Worst create(s)/sec = 6026.30 ( 0.000165939 seconds/op)
>
> write: Files = 8 Total Time = 0.001550436 seconds
> Avg write(s)/sec = 5159.84 ( 0.000193805 seconds/op)
> Best write(s)/sec = 6442.86 ( 0.000155210 seconds/op)
> Worst write(s)/sec = 4148.67 ( 0.000241041 seconds/op)
>
> close: Files = 8 Total Time = 0.000099897 seconds
> Avg close(s)/sec = 80082.18 ( 0.000012487 seconds/op)
> Best close(s)/sec = 127100.12 ( 0.000007868 seconds/op)
> Worst close(s)/sec = 27060.03 ( 0.000036955 seconds/op)
>
> stat: Files = 8 Total Time = 0.000183582 seconds
> Avg stat(s)/sec = 43577.18 ( 0.000022948 seconds/op)
> Best stat(s)/sec = 91180.52 ( 0.000010967 seconds/op)
> Worst stat(s)/sec = 10433.59 ( 0.000095844 seconds/op)
>
> access: Files = 8 Total Time = 0.000101566 seconds
> Avg access(s)/sec = 78766.27 ( 0.000012696 seconds/op)
> Best access(s)/sec = 144631.17 ( 0.000006914 seconds/op)
> Worst access(s)/sec = 23172.95 ( 0.000043154 seconds/op)
>
> chmod: Files = 8 Total Time = 0.000674963 seconds
> Avg chmod(s)/sec = 11852.50 ( 0.000084370 seconds/op)
> Best chmod(s)/sec = 33554.43 ( 0.000029802 seconds/op)
> Worst chmod(s)/sec = 3153.61 ( 0.000317097 seconds/op)
>
> readdir: Files = 4 Total Time = 0.000214100 seconds
> Avg readdir(s)/sec = 18682.87 ( 0.000053525 seconds/op)
> Best readdir(s)/sec = 38479.85 ( 0.000025988 seconds/op)
> Worst readdir(s)/sec = 7681.88 ( 0.000130177 seconds/op)
>
> link: Files = 8 Total Time = 0.000689983 seconds
> Avg link(s)/sec = 11594.48 ( 0.000086248 seconds/op)
> Best link(s)/sec = 16131.94 ( 0.000061989 seconds/op)
> Worst link(s)/sec = 4928.68 ( 0.000202894 seconds/op)
>
> unlink: Files = 8 Total Time = 0.000451565 seconds
> Avg unlink(s)/sec = 17716.17 ( 0.000056446 seconds/op)
> Best unlink(s)/sec = 25575.02 ( 0.000039101 seconds/op)
> Worst unlink(s)/sec = 6944.21 ( 0.000144005 seconds/op)
>
> delete: Files = 8 Total Time = 0.001273155 seconds
> Avg delete(s)/sec = 6283.60 ( 0.000159144 seconds/op)
> Best delete(s)/sec = 7307.15 ( 0.000136852 seconds/op)
> Worst delete(s)/sec = 4346.43 ( 0.000230074 seconds/op)
>
> Hope this helps,
> -Kums
>
>>>> "Priya Muralidharan"
<Priya_Muralidharan@symantec.com> 7/20/2006
> 4:40 PM >>>
> Hi Kumaran,
>
> Have you tested Lustre using fileop. Is there any documentation
> present on using Fileop? can you specify the folder where you want to
> run Fileop?
> or will I have to copy Fileop to /mnt/lustre and try it there?
>
> Thank You
> Priya
>
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss@clusterfs.com
> https://mail.clusterfs.com/mailman/listinfo/lustre-discuss
>